PassMyEssay Claude watermark remover and AI checker — strip the invisible characters and the phrasing that make Claude text read as machine-written so your draft sounds like you wrote it.
Paste your Claude output and hit Clean. Hidden Unicode, Artifact markup, and the Claude cadence — gone in seconds, meaning intact.
Input
Input
A normal paste out of Claude, with the invisible characters rendered so you can actually see them — and the phrasing anyone who uses Claude recognises on sight.
Happy to help — here's a breakdown. Remote work has reshaped how teams zwspoperate.nbspThat said, it's worth noting that the picture is more nuanced than it first appears. This isn't just a change in where people work—it's a change in how organizations think about trust.zwsp Let me know if you'd like me to expand on any of this.
Every paste out of Claude carries some of this. Check your own text
No install, no signup to try it, no file upload. Paste, clean, copy.
From the chat window, an Artifact, or the desktop app. Everything invisible that came with it — zero-width spaces, non-breaking spaces, stray markup from the rendered document — comes along too.
Every copy artifact is stripped in one pass, and the Claude phrasing gets rewritten — the acknowledging opener, the stacked caveats, the bolded-bullet scaffolding. Your facts, quotes, and citations are held exactly as you wrote them.
Paste into Word, Google Docs, Gmail, or WordPress with nothing hidden in the byte stream and nothing in the wording that names the model.
A humanizer that rewrites machine-sounding prose into your voice, and a detector that shows you what's flagged so you can correct it yourself. Watermarks and hidden characters are cleaned automatically along the way.
When you copy text out of Claude, you carry two things away with you that you didn't ask for. A set of characters that never appear on your screen, and a set of writing habits that appear very clearly to anyone who reads the result. Together those layers are what people mean by a Claude watermark.
The first layer is a copy-path artifact. Claude's interface — and Artifacts especially — renders markdown into real HTML, so a copy hands you the rendered document rather than the plain text you thought you were taking. The second layer is stylistic, and it is the one that costs you something. Claude has a distinctive voice: measured, hedged, generous with caveats. Deleting characters does nothing to it.
PassMyEssay is built for the second layer. The invisible characters get cleaned automatically on the way through, without you asking.
These live in the byte stream, invisible in every editor, and they travel with the document forever unless something strips them:
**bold** fragments and list markers that survived a plain-text paste and now sit in your document as literal asterisksAnyone who has used Claude for a week can name these from across the room:
Claude writes well, which is precisely the problem. Its prose is balanced, its paragraphs are the right length, and it qualifies its claims the way a careful person would. Nobody sustains that. Real writing under deadline has a rough patch, a sentence that runs too long, a paragraph that commits harder than the one before it. Prose that stays perfectly even-handed for eight hundred words is prose a reader starts to squint at.
The structural habits are louder still. A Claude answer wants to be a list with bolded lead-ins, and once it becomes an essay those bullets flatten into paragraphs that still open the same way. Every section starts by naming what it's about. Every claim gets a counterweight in the following clause. Read three of them in a row and the shape gives it away before the vocabulary does.
And then there's the third case, the one people find most upsetting. Detection tools score writing style, not authorship. Careful, formal, evenly-structured prose — exactly what school trains you to produce — scores like a machine. Students who wrote every word themselves get flagged. Writers working in a second language get flagged at higher rates still, for the crime of being consistent.
Paste and hit Clean. Every invisible character goes in a single pass, and the Claude phrasing is rewritten into prose with actual rhythm — the acknowledging opener cut, the stacked caveats resolved into a position, sentence lengths varied the way human writing varies. Your claims, numbers, dates, quotes, and citations are held fixed the whole way through.
Sometimes you don't want anything rewritten for you. AI Phrase Detector highlights the exact sentences that read as machine-written, tells you why each one was flagged, and leaves the editing to you. Nothing changes unless you change it. Most people find this is the mode that teaches them something — after a few passes you start catching the Claude cadence in your own drafts before the tool does.
Deleting a hidden character is a lookup. Anyone can do it, and we do it automatically without being asked. Deleting Claude's voice is a different problem, because the tells aren't words on a blocklist — they're structural. You can delete every "that said" in the document and the paragraph will still move exactly like Claude wrote it.
Fixing that means understanding the sentence. Which claim is load-bearing. Which qualifier is doing real work and which one is just hedging out of habit. Where the rhythm has gone flat and how to break it without losing the argument. There is no rule table for that, which is why our humanizer is model-driven and why it reads your draft in full before it changes a word.
We'll be plain about the limit. We won't promise you a specific score on a specific detector, because detectors change weekly and anyone promising that number is selling you something. What we'll promise is writing that sounds like a person wrote it.
Make it the last thing you do before sending. Copy artifacts accumulate every time you paste out of a chat window, so cleaning once at the end beats cleaning as you go.
Two layers of Claude residue: the characters you can't see, and the voice everyone can.
Zero-width spaces (U+200B), word joiners, non-breaking spaces, and the rendered markup that rides along when you copy out of an Artifact. You can't see any of it. Word, Google Docs, and every parser can.
"Happy to help." "That said." "It's worth noting." "This isn't just X — it's Y." The bolded lead-in on every bullet. The tells anyone who uses Claude clocks in three seconds, rewritten out.
AI Phrase Detector highlights the exact sentences that read as machine-written and tells you why each one was flagged. You correct them yourself — nothing changes unless you change it.
Numbers, dates, names, quotes, technical terms, and citations are held fixed through every clean. We change how it reads, never what it claims.
No extension to install, no signup wall to test it, no file upload. Paste, hit Clean, copy the result. The whole thing takes about ten seconds.
It doesn't need to know which model you used. Claude, ChatGPT, Gemini, grok, Llama, and Copilot all leave the same class of copy artifacts and lean on their own stock phrasing. All handled.
What they are, what Anthropic does and doesn't do, and how to get all of it out of your text.
Two separate things that get called by the same name. The first is technical: invisible Unicode — zero-width spaces, word joiners, non-breaking spaces — that comes along when you copy out of the Claude interface or out of an Artifact. The second is stylistic: the phrasing and structure Claude falls into by default. The first is invisible to you and obvious to a parser. The second is invisible to a parser and obvious to anyone who has used Claude for a week.
Anthropic has not published a text watermarking scheme for Claude, so there is no known hidden signature deliberately stamped into the words. What does happen is that copying from the web app drags formatting characters and markup along with the text, and that Claude's prose has a strong statistical fingerprint. Neither is a tracking system. Both are detectable, which is the part that actually affects you.
Paste your text into the workspace at the top of this page and hit Clean. Every invisible character is stripped in one pass, and the Claude-specific phrasing gets rewritten. Copy the result out. No signup to try it, no extension, no file upload.
Artifacts render markdown into real HTML, so copying from one gives you the rendered document rather than the plain text — non-breaking spaces where the layout needed them, zero-width characters holding inline formatting together, and sometimes styling attributes that ride straight into your CMS. Copying from an Artifact into Word or WordPress is where people notice the damage first.
The opener that acknowledges before it answers — "Great question", "I'd be happy to help with that", "Let me break this down". The bolded lead-in on every bullet in a list. "It's worth noting", "That said", "Importantly". The correction-shaped sentence: "This isn't just X — it's Y." And the closing offer, "let me know if you'd like me to expand on any of this", which people paste into a document without ever reading it back.
Claude hedges more and qualifies more. It stacks caveats mid-paragraph, it likes long dependent clauses, and it reaches for a nuance-signalling aside where ChatGPT would just assert. The result reads as more thoughtful, which is exactly why it is easy to spot: real writing under deadline is rarely that even-handed for that many paragraphs in a row.
Yes, in small expensive ways. A zero-width space inside a word makes find-and-replace miss it. A non-breaking space wraps differently and ruins justified layouts. Markup copied out of an Artifact crashes Gutenberg blocks. A citation manager fails to match a reference it should have matched. None of it is dramatic and all of it eats an afternoon.
No. Numbers, dates, names, quotations, technical terms, and citations are held fixed through every pass. We change how it reads, never what it claims.
Clean does the work for you: strips the characters and rewrites the phrasing in a single pass. AI Phrase Detector does none of it — it highlights the exact sentences that read as machine-written, explains why each was flagged, and leaves the editing to you. Use Clean when you want it handled, AI Phrase Detector when you want control over every sentence.
Yes, and it doesn't need to know which one you used. Every model in the family shares the same house style, and every version of the web interface copies the same class of formatting characters. The same goes for text pulled from the Claude API, the desktop app, or a Claude-powered tool built by someone else.
Yes. The invisible-character problem is identical across chat interfaces, and the stock-phrasing problem differs only in dialect. If your draft is a mix of models — which most drafts now are — paste the whole thing.
Your text is processed and returned, not retained. It doesn't go into a repository, it isn't sold, and it isn't used to train anything. Checking your own essay here can't come back to haunt you later.
Because they measure style, not authorship. A detector looks at how much your sentence length varies and how predictable your word choices are. Careful, formal, well-organised writing scores similarly to AI on both counts — which is why genuine student work gets flagged constantly, and why non-native English speakers get flagged at higher rates.
No, and be sceptical of anyone who does. Detectors update constantly and their output is probabilistic. What we can promise is text that comes out clean in the byte stream and reads like a person wrote it — the part we actually control, and the part that matters whether or not a scanner is ever involved.
Not on work that's genuinely yours. Stripping invisible characters is hygiene, and tightening phrasing on something you researched and argued is what an editor does. Submitting work you didn't write is a different thing, and no tool makes that honest — including this one. Follow your institution's or employer's policy on AI assistance.
You can try it without paying and without a card — the workspace on this page shows you the output quality on a short slice of text. Paid plans unlock full-length documents, starting at $7.99 for a Weekly Pass.
Strip the invisible watermark characters and the phrasing that gives it away — or just see what's flagged and fix it yourself. Free to try, no card required.
Try a preview without paying. Cancel anytime once you upgrade.